Re: [PATCH v2 1/6] KVM: SVM: Use maxphyaddr in emulator RAX check for VMRUN/VMLOAD/VMSAVE

[Sean Christopherson <seanjc@google.com>]

On Thu, Mar 12, 2026, Yosry Ahmed wrote:
> > diff --git a/arch/x86/kvm/emulate.c b/arch/x86/kvm/emulate.c
> > index c8e292e9a24d..74df977a38ca 100644
> > --- a/arch/x86/kvm/emulate.c
> > +++ b/arch/x86/kvm/emulate.c
> > @@ -3867,18 +3867,10 @@ static int check_svme(struct x86_emulate_ctxt *ctxt)
> >         if (!(efer & EFER_SVME))
> >                 return emulate_ud(ctxt);
> >
> > -       return X86EMUL_CONTINUE;
> > -}
> > -
> > -static int check_svme_pa(struct x86_emulate_ctxt *ctxt)
> > -{
> > -       u64 rax = reg_read(ctxt, VCPU_REGS_RAX);
> > -
> > -       /* Valid physical address? */
> > -       if (rax & 0xffff000000000000ULL)
> > +       if (ctxt->ops->cpl(ctxt))
> >                 return emulate_gp(ctxt, 0);
> >
> > -       return check_svme(ctxt);
> > +       return X86EMUL_CONTINUE;
> >  }
> >
> >  static int check_rdtsc(struct x86_emulate_ctxt *ctxt)
> > @@ -3984,10 +3976,10 @@ static const struct opcode group7_rm2[] = {
> >  };
> >
> >  static const struct opcode group7_rm3[] = {
> > -       DIP(SrcNone | Prot | Priv,              vmrun,          check_svme_pa),
> > +       DIP(SrcNone | Prot | Priv,              vmrun,          check_svme),
> >         II(SrcNone  | Prot | EmulateOnUD,       em_hypercall,   vmmcall),
> > -       DIP(SrcNone | Prot | Priv,              vmload,         check_svme_pa),
> > -       DIP(SrcNone | Prot | Priv,              vmsave,         check_svme_pa),
> > +       DIP(SrcNone | Prot | Priv,              vmload,         check_svme),
> > +       DIP(SrcNone | Prot | Priv,              vmsave,         check_svme),
> >         DIP(SrcNone | Prot | Priv,              stgi,           check_svme),
> >         DIP(SrcNone | Prot | Priv,              clgi,           check_svme),
> >         DIP(SrcNone | Prot | Priv,              skinit,         check_svme),
> > diff --git a/arch/x86/kvm/svm/svm.c b/arch/x86/kvm/svm/svm.c
> > index e6691c044913..e1223c07593b 100644
> > --- a/arch/x86/kvm/svm/svm.c
> > +++ b/arch/x86/kvm/svm/svm.c
> > @@ -2294,7 +2294,7 @@ static int gp_interception(struct kvm_vcpu *vcpu)
> >                                 EMULTYPE_VMWARE_GP | EMULTYPE_NO_DECODE);
> >         } else {
> >                 /* All SVM instructions expect page aligned RAX */
> > -               if (svm->vmcb->save.rax & ~PAGE_MASK)
> > +               if (!page_address_valid(vcpu, svm->vmcb->save.rax))
> >                         goto reinject;
> 
> Final observation (hopefully), this check needs to be moved to the
> VMRUN/VMLOAD/VMSAVE interception functions.

Gah, yeah.  I noticed that when initially typing up my response, but lost track
of it when I got distracted by all the emulator crud.

> As kvm_vcpu_map() failures will stop injecting #GP, we still need to handle
> the case where allow_smaller_maxphyaddr is used and the GPA is illegal from
> the vCPU's perspective but not the host.

allow_smaller_maxphyaddr is irrelevant.  My read of the APM is that the intercept
has priority over the #GP due to a bad RAX.  So with vls=0, KVM needs to check
RAX irrespective of allow_smaller_maxphyaddr.